Participants performed a two-alternative forced-choice task on a foveally presented stimulus that could vary on a subset of binary perceptual features, such as color (red, green), shape (diamond, square), size (large, small), topology (open, closed), and location (up, down). Unbeknownst to the participants, we manipulated the statistical informativeness of an additional feature that was not part of the task, such that this feature always predicted the correct response in one condition (the predictive condition) but not in the other condition (the baseline condition). Because the cognitive system is known to exploit statistical stimulus-response contingencies automatically, performance was expected to be better in the predictive than in the baseline condition.
We embedded these predictive and baseline conditions into two different tasks, which we thought would induce different cognitive-control states. The control task included instructions intended to emphasize the need for top-down control: Participants were instructed to classify the stimulus according to a feature-conjunction rule (e.g., size and topology: left response key for large and open or small and closed shapes, right response key for small and open or large and closed shapes). The automatic task included instructions intended to deemphasize the need for control: Participants were instructed to classify the stimulus according to a single feature (e.g., shape: left response key for a diamond and right response key for a square). In the automatic task, the features were mapped consistently on responses and thus allowed automatic visuomotor translation. In contrast, the stimulus-response mapping in the control task required the attention-demanding integration of two features before the response could be determined.
As expected, the predictive feature improved performance when participants performed the task automatically. Counterintuitively, however, the predictive feature impaired performance when subjects were performing the exact same task in a top-down, controlled manner.Their abstract:
In order to engage in goal-directed behavior, cognitive agents have to control the processing of task-relevant features in their environments. Although cognitive control is critical for performance in unpredictable task environments, it is currently unknown how it affects performance in highly structured and predictable environments. In the present study, we showed that, counterintuitively, top-down control can impair and interfere with the otherwise automatic integration of statistical information in a predictable task environment, and it can render behavior less efficient than it would have been without the attempt to control the flow of information. In other words, less can sometimes be more (in terms of cognitive control), especially if the environment provides sufficient information for the cognitive system to behave on autopilot based on automatic processes alone.